This entry starts with analytical issues associated with the concept of sentence and then turns to more general issues.
Traditionally, the sentence is the domain within which purely syntactic relations, constituent structure, and grammaticality are defined. It is the one independent syntactic entity. It expresses a predication and is often regarded as the linguistic vehicle for the expression of a propositonal attitude and thus the performance of a speech-act.
That much would generally be held to be true cross-linguistically, though how it is realized in syntax and morphology differs from language to language. We focus here on English. Traditionally, English sentences consist of one expression functioning as subject and another as predicate, the latter centered on a verb which must be finite (i.e., inflected for tense/subject agreement). What further expressions must figure in the predicate depends on the type (subcategory) of the verb: none with intransitive verbs (Tom [laughed]), one with transitives (Tom [ate – the pies]), two with ditransitives (Tom [gave – them – the pies]), for example. Sentences are declarative (all of those examples), interrogative (What did Tom eat?), exclamative (What a long meeting that was!), or imperative (Eat those pies!), where the subject is not overt because understood as referring to the addressee.
Sentences may contain sentence-like constituents, traditional subordinate clauses. In these, the subject may be nonovert because understood (compare Tom wants [Anna to go] with Tom wants [to go], where Tom is the subject of to go), and the verb need not be finite (to is the infinitive particle). Subordinate clauses contribute to the speech-act performable at sentence level but don’t in themselves allow for the performance of a speech-act. A subordinate interrogative clause, for example, is not a question (Tom asked [who ate the pies]). Simple sentences consist of one clause (sentence equals clause here). Complex sentences consist of a main clause and any number of subordinate clauses. (See Burton-Roberts 2010, a basic introduction to English sentence analysis, and Huddleston and Pullum 2002, a comprehensive grammar, in which “sentence” is abandoned in favor of “clause.”)
In what follows, we trace how sentence has fared in Chomskian generative grammar (CGG, henceforth). Noam Chomsky (1957) was traditional in taking S as the symbol to be defined by any grammar – the initial symbol and thus the root node in any phrase structure tree. In defining S for a language L, a grammar was said to generate the sentences of L and thereby generate L. The definition consisted of successive rewrite rules, the first being [S → NP-VP], where NP (noun phrase) functions as subject and VP (verb phrase) as predicate, though predicate is seldom used in this context (see predicate and argument). Chomsky (1965) adopted an alternative initial rule, which informed subsequent developments: [S → NP-AUX-VP]. Here, “AUX” is the locus of tense/agreement, as in S[NP[Tom] AUX[has] VP[eaten the pies]]. Subordinate clauses were treated as embedded sentences. To accommodate clause-introducing complementiser expressions (e.g. that/whether/when in Anna knows that/whether/when Tom laughed), an extension of S was introduced: S’ (“S-bar”). This was defined/expanded by the rule [S’ → Comp-S], designed to capture the fact that [that/whether/when [Tom laughed]] is in some sense sentential (clausal), while distinguishing it from the basic clause itself (Tom laughed). In terminology adopted later, S’ is a projection of S. This projection idea was developed in x-bar theory, with consequences for S as a formal category.
Pregenerative (Bloomfieldian) linguistics distinguished between endocentric and exocentric structures. Endocentric structures are headed, centered on a constituent of the same category as the whole structure. Thus, Tom’s summary of the argument is an NP because centered on the noun summary. In other words, the NP is a projection of its head, N. Now, thought of as constituted as either [NP-VP] or [NP-AUX-VP], S is a category distinct from the categories of its constituents and, hence, not a projection of any of them. S would therefore seem to be headless (exocentric). By contrast, X-bar theory (initiated in Chomsky 1970) makes the constraining assumption that all categories are endocentric and have the same three-level projective structure (XP, X’, X). With S treated as the (exocentric) root node, we miss the parallelism between the above NP and the sentence Tom summarized the argument. If summary is head of the NP, the inflected verb summarized should be the head of the corresponding sentence. This suggests that the root node is not S but a projection either of V or, given NP-AUX-VP, of the tense/agreement inflection.
Sentence is therefore abandoned as a formal category, at least in CGG. An inflection phrase system (IP, I’, I) is posited, with the traditional subject treated as the specifier of IP (Spec, IP). Furthermore, this “Infl” system is itself embedded within a complementizer phrase system (CP, C’, C). This goes for all clauses, including main clauses (e.g., CP[who C’[ C[did] IP[he marry t]]]). The root node, therefore, is CP.
Although S(entence) has been abandoned as a formal category in CGG, the term sentence is still widely used even in that context, but informally. This is how it is used in what follows, where we turn to more general questions.
How does the structural notion of sentence (however analyzed) map onto speech-acts and speakers’ utterance behavior? This can be seen as a question of competence versus performance, interfacing with pragmatics (see also semantics-pragmatics interactions).
It is uncontroversial that speakers utter words, one after the other. But do speakers utter sentences? Strings of uttered words can be structurally ambiguous (He-watched-the-man-with-the-telescope). However, as described, a sentence has – indeed is – a unique structure (generated, we assume, by a mentally represented grammar). A structure, as such, cannot be structurally ambiguous. This suggests that word strings can be structurally ambiguous because they don’t, in fact, have syntactic structure. Arguably, this is why word strings require parsing. Parsing, on these terms, is a matter of putting a (structureless) word string into correspondence with a sentential structure. Structural ambiguity in a word string, rather than being a matter of “having more than one structure,” occurs when the string can be put into correspondence with more than one uniquely structured sentence. This suggests that we do not utter sentences as such.
Nevertheless, it is generally assumed, if only informally, that uttering certain word strings counts as uttering a sentence. On that assumption, sentence is a somewhat ambiguous term, applicable both to mind-internal structures and to what can be uttered/heard in speech.
Even allowing that sentences can be uttered, sentence and utterance are not isomorphic. For example, utterances may include parenthetical elements, some of which are not clearly constituents of sentence structure (“It’s – I don’t know – let’s see now – about twenty miles”) and some whose status as sentence constituents is controversial – for example, appositive relative clauses. Notice that in uttering “Tom, who eats all the pies, is getting fat,” we perform two (assertive) speech-acts. (See Burton-Roberts 2005 on parentheticals.)
Furthermore, speech-acts performable in uttering sentences are performable by nonsentential utterances. “Yes” in answer to “Did Tom laugh?” communicates what’s communicated by “Tom laughed.” Is the utterance of yes the utterance of a sentence? Possibly – if yes is a sentential pro-form (a pro-sentence, parallel with pronoun and the pro-VP do so). The question arises more urgently with nonsentential utterances that are clearly not pro-sentences: “Possibly” (just used), “What a day!” “Ready?” and “Sleep” in reply to “What did you do today?” A syntactic approach, appealing to ellipsis, would analyse them as utterances of sentential structures (e.g., Are you ready? or What I did today was sleep) in which words, or at least their phonological features, are deleted. A nonlinguistic, pragmatic approach would treat them as utterances of just the words heard, explaining what they communicate by appeal to distinct conceptual structures in “mentalese” (see language of thought). (See Stainton  for discussion and references.) The issue arises even with utterances not generally regarded as elliptical. Is “Eat!” the utterance of an imperative sentence or just a verb? “It’s raining” generally communicates that it’s raining here (where the speaker is). But does the sentence uttered include a covert location variable whose value is given in the context of utterance, as a matter of sentence semantics, or is the location necessitated (and supplied) independently by conceptual structure? (See Stanley 2000; Recanati 2002.)
At issue is the relation of sentences to utterances, on the one hand, and thoughts, on the other. How sentence is understood depends on how sentences are felt to be related to – and distinguished from – thoughts and utterances.
A self-contained, grammatical, linguistic unit consisting of one or more words syntactically and semantically related to each other such that...
[13 century] ‘Complete grammatical unit’ is a comparatively recent meaning of sentence , which only emerged in English in the 15th century....
Within a sentence, constituents may appear in the local context of the items they stand in a grammatical relation with or they may move to other...